In MOEA/D-EGO algorithm, when there are too many modeling sample set elements or the population scale is large , it will lead to a long computation time. In order to reduce the run time of the MOEA/D-EGO algorithm, this paper parallelizes both the modeling process and the population optimization process. considering the experimental conditions, this paper uses the master-slave parallel model which adds the task to the main process in the condition of fully considering the efficiency of computer resources and load balance. The main process not only assigns computation task, distributes data, configures algorithm, collects the computation results, but also participates in the task of child process and complete the same amount of computation task as child process. The experimental result shows that the paralleled MOEA/D-EGO algorithm can effectively solve the multi-objective optimization problem, and can significantly shorten the running time of the algorithm.%在MOEA/D-EGO算法中，当建模样本点集合元素太多和种群规模较大时，会导致算法运行时间过长。为了减少 MOEA/D-EGO 算法的运行时间，文章对 MOEA/D-EGO 算法的建模过程和种群优化过程同时并行化。在综合考虑实验条件限制的情况下，使用了基于主从式的并行模型，模型在充分考虑计算机资源的使用效率与负载均衡等因素下，增加了主进程的任务，主进程不仅需要为子进程分配计算任务、分发数据、进行算法配置、收集子进程返回的计算结果，还需要参与子进程的任务，完成与子进程相当量的计算任务。实验结果表明文章的并行MOEA/D-EGO算法能有效求解多目标优化问题，且能够大幅缩短算法运行时间。
展开▼
